What's Up with Tech?

Turning AI Hype Into Enterprise ROI

• Evan Kirstel

Share episode

Interested in being a guest? Email us at admin@evankirstel.com

Hype is loud, but outcomes win. We sit down with Smartling CEO Bryan Murphy to explore why the center of gravity in AI is shifting from raw models to the application layer where workflows, data, and governance translate into real business value. If you’ve been stuck in pilot purgatory, you’ll hear how to move from experiments to production with clear ROI.

Bryan draws sharp parallels to earlier waves—the internet and cloud—and explains how today’s “chips and data centers” moment mirrors the 90s buildout, with a familiar follow-up: purpose-built apps that make the infrastructure usable. We dig into translation as a top-tier AI use case and unpack Smartling’s results: translating 3x more content at about 60 percent less cost and six times faster than traditional methods. The secret isn’t a flashy bolt-on; it’s re-architecting around an AI hub and agentic workflows that integrate directly with enterprise stacks.

We also challenge the “AI will kill SaaS” headline. Foundational models like GPT and Claude are treated as infrastructure, while the real differentiation comes from orchestration—selecting the best model for the job, routing data safely, learning from feedback, and guaranteeing quality. Bryan shares how governance protects brand voice at scale, from hallucination detection to automated language QA and custom-trained models honoring glossaries and style guides. With 7 billion words translated a year, reliability isn’t a feature—it’s the product.

Looking ahead, we explore multimodal translation across text, audio, and video, and a future where translation functions as a service embedded in everyday tools. For founders, Bryan offers practical advice: target a 10x problem, verify there’s real TAM, and build applications that customers will happily pay for. For enterprise leaders, the message is simple: stop chasing models, invest in orchestration, and pick solutions that measure and deliver outcomes.

If this conversation helps you think clearer about where AI value is headed, follow the show, share it with a colleague, and leave a quick review so others can find it.

Support the show

More at https://linktr.ee/EvanKirstel

SPEAKER_00:

Hey everyone. Fascinating chat today with the CEO of Smartling, a guy who knows a thing or two about turning hype into real enterprise ROI. Brian, how are you?

SPEAKER_01:

I'm doing great, Evan. Thanks for having me.

SPEAKER_00:

Well, thanks for being here. Before we dive in, perhaps introduce yourself and how would you describe SmartLing these days?

SPEAKER_01:

Sure. Well, to start, my name is Brian Murphy. I'm the CEO of Smartling, and we are an AI translation company. So our customers include companies like Apple, IBM, OpenAI, Anthropic, Disney, Pepsi, you name it. Lots of big global enterprises that use us to help create multilingual experiences that their customers love.

SPEAKER_00:

Fantastic. So let's talk big picture. You've been around the block uh a couple of times. We've seen the internet, the cloud, and now the AI revolution up close. What feels the same this time or or what's totally different in your opinion?

SPEAKER_01:

Yeah, you know, it's it's funny. Um this uh this feels a lot like uh the late 90s, you know, with the commercialization of the internet. Um back then, all the infest uh the initial investment was in infrastructure, right? Remember telecom, right? We had to do all that. Um and then um and and now uh the investment is in uh is in infrastructure, right? So data centers, chips, and uh building out the the models, the the uh the AI models. And I think that um these technology waves, I've been through at least three of them now, and they they tend to follow a similar pattern. First, the infrastructure gets built out, and then um and then business applications, or I should say not business, but applications start um getting built, right, to take advantage of that new infrastructure. So you can think of an example like Salesforce is a great example, right? So um back in the uh as cloud was being built out, they didn't invent cloud, but they made um they made their application, they made it accessible and transformable for the business users. So I think that the same thing is happening right now. Foundational models are the new infrastructure, but the real value is is beginning to get created up the stack. Um companies like Smartling that make AI useful, measurable, and ROI-driven um and real workflows with purpose-built applications is the way I describe it.

SPEAKER_00:

Yes, well, if if there is a killer app, you certainly have one. Um so all the hype and excitement around infrastructure is great for geeks like me. Uh, I I'm fascinated by it. But uh, do you see more and more of the value in AI shifting to the app layer and and from the sort of model layer? I yeah, I do.

SPEAKER_01:

I do. I think, you know, um, once again, like if we go back to the commercialization of the internet, I mean, when we were building out our first e-commerce companies, we had to build everything. I mean, there wasn't a shopify, right? We had to build our own payment systems, we had to build our own CRMs, we had to build our own pin, you know, like catalogs, everything, right? Um, and then companies come along and they build those purpose-built applications to solve real, real business problems. And I think that's happening here. We see um, you know, as the foundational models, um, they're they're the pace of in the pace of uh improvement is is slowing, right, on them. And companies now you're beginning to see companies like um you know, like Harvey and Cursor and Smart Link creating like these these these applications that solve very um specific problems. And so translation is probably one of the top five use cases of AI out there, right? There's uh we know that it's uh AI is great at content generation, it's it's really having a big impact on generating code. And translation is definitely one of those things. And when we look at that impact, um, you know, our customers are able to translate uh 3x more content for about 60 percent, yeah, about 60 percent less cost and six times faster than traditional. So that's the impact of AI on translation.

SPEAKER_00:

That is fantastic. And you know, what are traditional SaaS companies getting wrong about AI or or their what what is what about their approach would do you think they're they're not getting right?

SPEAKER_01:

Um I think that probably one of the one of the big challenges that traditional SaaS companies um have, or one of the challenges they have is mindset. So they tend to the the thinking is typically to um for legacy SaaS company to bolt AI features onto old workflows. And that doesn't work real great. We've all kind of had that experience. The uh the real opportunity is um, and the requirement is I think you have to rethink the entire experience. Um, what data do you collect, how users interact with the application, how output gets measured. And what that meant for us was really re-architecting SmartLink with our AI hub, uh generating agentic workflows and um different ways of integrating to our customers' tech stack. So really completely transforming that experience. And I think it like this was painful. Like we had to redesign uh our application in order to make it work in this new mode. It wasn't it wasn't a bolt-on, it was a reimagination of how it works.

SPEAKER_00:

Fantastic. And there's you know lots of talk about the death of SaaS uh by AI or or the models themselves, which seems pretty overstated, uh, of course. Um, but what what are the risks to traditional SaaS models, maybe to your model, not just from competition, but from the models themselves?

SPEAKER_01:

Yeah, I think there definitely is a risk. I mean, if you if you go back to cloud, right? You remember um uh, you know, I mean, there's companies like SIBL and Oracle, and and they're still, and and Oracle obviously is around very successful, but that was an incredibly painful prop uh process for them when that old model of selling software licenses, right?$2 million deal, like was the way they thought about the world, right? Then all of a sudden companies like Salesforce and HubSpot come along and they're like, hey, you know what, we're gonna charge you per seat per user per month, um, and this cloud-based, right, not not on-prem uh sort of thing. And it was like a complete shift from doing business and the applications and all of that. And I think that the risk is for legacy SaaS companies that don't make that shift, that they will be supplanted by um by AI first uh, I'll call them SaaS companies or application, maybe we'll call them application companies, right? So I don't think that like so, for example, I do view like uh GPT, we I mean GPT, Anthropic, um, you know, Cloud, all at Gemini, they're all these are all partners of ours. They're they're tremendous technology, but I do view them as infrastructure rather than applications. They're very good at doing like kind of like anything and everything, right? But like in order to get really good um at doing something, you have to focus on it.

SPEAKER_00:

Oh, really well said. And there's lots of talk in the media and elsewhere on um AI successes, but also AI failures, lots of projects not succeeding, lots of trials, experimentation, science projects. You're well past the science project phase. You're in high-volume production with who's who in the enterprise. Uh, how can enterprise teams make sure their pilots kind of deliver real ROI and not just you know headlines or splashy, you know, demos?

SPEAKER_01:

Yeah, we're we're well past that. We uh think it was interesting. Just uh yesterday I was checking and um nearly nearly 80% of the translation that we deliver is now powered by AI, right? As of Q3, which is um pretty pretty remarkable when you think about it. And we're probably one of the fastest growing translation companies in the$30 billion market. So um it's not it's not insignificant. Um I think I see, you know, we you know, there's a famous MIT study that recently came out that said uh that uh indicated that 95% of enterprise AI projects were were failing, right? And I think that one of the problems or one of the reasons for that is that companies treat tend to have a tendency to treat AI like a science project, you know, lots and lots of pilots, uh, you know, 20 plus pilots, lots of excitement, but without a specific business outcome in mind. So I I do think, and I also do think that you know, Jack Welsh famously said, you know, um to outsource, you know, everyone's got a everyone's got a front office. It's always a good idea to back so outsource your back office to someone's front office, right? And with with the AI being having the potential to do everything, and by the way, incredible pressure from investors, board of directors, on CEOs to produce results with AI. There's this incredible pressure within enterprises to use AI, right? Um, what I've found is that, and and so, and we're still early in the cycle, right? So there's not a lot a lot of applications for them to do that. So they're being forced to sort of experiment. I think over the next 24 months, you're gonna start seeing, and you are starting to see these really great AI applications coming out that are having huge impacts on their enterprise customers. And you're gonna see this migration away from this sort of experimentation to, hey, you know what? Um, that's a great solution. Let's use that. I actually have a funny example of I I actually did this ourselves. I'm one of that, we're part of that 95%. I was um frustrated with the fact that we didn't have AI uh for our customer service, right? So someone could type into a chat box and chat box and get an answer. I'm like, we're an AI company, let's do this. So we um we actually I directed my engineering team to build that solution. And uh 90 days later, we were struggling getting the uh the results we wanted to, and we um went out to uh a customer partner of ours called Intercom, and they have this um AI application called Fin for customer service, and it worked great. Plugged it right in, did everything we wanted to do. But that's like sort of my point. Like I was being forced as a CEO to start experimenting and building this stuff internally because there wasn't this available solution. Now these available solutions are starting to pop up, and I think you're gonna see enterprises gravitating to those purpose-built applications and really get tremendous results.

SPEAKER_00:

Great point. And there's some indication that model innovation is flowing down, or at least you know, going through a bit of a trough uh uh given the existing technologies. Um we've seen this explosion of innovation, but where should B2B software companies focus their energy next as we you know consume all these models and get to understand how they work?

SPEAKER_01:

Yeah, you know, the models are still improving, albeit at a slower pace, right? So that the first iterations like were big step changes, and now you're getting into a little bit more of this granularity in terms of improvement. And um, they're, you know, they're they're they're they're they're yeah, they're just not improving as quickly as they were before. And so I would stop chasing models, right? Which one's gonna be better, and start focusing on building orchestration, right? So uh as foundational models become more like the cloud, powerful but largely invisible, the real differentiation is how you connect them to solve your customers' problem, your data, and your workflow. So every Sirius SaaS company should be focusing on building its own AI orchestration layer. Um, we do that with we we do that, our ours is called AI Hub, and we use that to route data, select the best model, learn continuously from customer feedback. And I think that I think the companies that uh that will win this decade uh will be the ones that make AI usable for their customers.

SPEAKER_00:

Fantastic. And you have quite a balancing act, it seems that SmartLink AI innovation on one hand with customer trust. That must be huge, hugely important, and translation quality. How do you think about juggling all those balls when things are moving so fast?

SPEAKER_01:

Well, it's it's critical. Um, you know, when you think about what SmartLink does, we're we're helping them transform their content uh into global experiences, right? And that's all about brand. And so you can't have missteps, right? Um, you know, you're creating source content, uh translating it into 15 different languages. You need to know what you just said. You don't have room for hallucination, right? Uh a mistranslation has created many, many of a brand headache or customer uh relationship headache for CEOs around the world. So we take that very seriously. So a big part of what we do is we obviously integrating to the existing Tech Act for automation, we automate 99% of our customers' translation. We then um we then build custom trained models uh for translation for each of the companies. So it matches, it's so it contains their translation, memory, glossary, style guide, uh brand voice, right? So if I'm uh any one of these companies, any one of our customers translating to a multitude of different languages, it's going to carry that brand voice in any of those countries and match the tone and style of voice that's required, right? So that's that's a big part of what we do. But then very importantly, what we do is on the back end of that, the governance. Um so we've got significant governments, governance uh policies in place. Um, we have hallucination detection and mitigation. Um, we have automated language quality assurance. So that our customers, yeah, this is all important. So when you know, we have got we we translate 7 billion words a year for our customers. Well, right. So um at an extremely high scale and in an automated way, uh we need to be able, our customers need to be able to rely on us that that content is coming out on brand and with very, very high quality.

SPEAKER_00:

Fantastic. Well, this isn't a tech uh deep dive, but I would be curious maybe you can give us a peek into the future, your roadmap, where things are headed. I can imagine with translation, there's things like real-time audio, real-time video translation, and so many use cases, so many applications. What is how will this evolve over the next one, two, five years?

SPEAKER_01:

Yeah, I think right now, if you look at where we are, we're right now our customers as a result of using Smartling AI translation are able to translate 3x more content for 60% less cost, six times faster than traditional, right? That's a massive impact right there. Where we see that going over the next 24 months is uh like as you indicated, multi-modality, as we call it, right? So that's uh video, audio, et cetera, added into the the traditional use cases. Um and really think about translation as a service, right? So that's sort of like our end game where um users within the enterprise that need to have multilingual content, they're able to create that multilingual content uh seamlessly within their platforms and uh be assured of being able to produce on-brand, high-quality, uh translated localized content.

SPEAKER_00:

Yeah, that's fantastic. I I did play with an app that translates, you know, your voice, uh, a consumer app in you know eight different languages, including your lip thinking and your you know, tonality, your your tone of voice. And it was pretty shocking to see. I shared it with friends, one was Hindi, one was French, and they said, Yeah, it's you, it's it's me, it's in my voice. I just can't imagine the possibilities. And speaking of a possibilities, you know, there's so many startup founders getting into this space, not just your space, but AI. Uh, what's your advice to them as starting, you know, if they want to build an AI app layer company? What's your word? What are your words of wisdom uh and uh guidance?

SPEAKER_01:

I think number one is identify a problem that you can fix like by 10x, right? It's gotta be uh like incremental is really tough to do. Uh you want to really have a huge impact. And then when you build that product, make sure that you've got enough tan, uh total addressable market. A lot of time, uh uh, you know, sometimes people are I always say, are you are you developing uh a feature or a product? Right? That's one thing. And then B, is the um is the total addressable market large enough? In other words, are willing people willing to are enough people willing to pay enough money for it that you can create a$200 million business?

SPEAKER_00:

Well, great advice and congratulations on all the success, onwards and upwards. So much more work to be done. Thanks, Brian.

SPEAKER_01:

It was my pleasure. It was great talking with you, Evan. Thank you.

SPEAKER_00:

Yeah, thank you. And thanks everyone for listening, watching, taking on our TV show, techimpact.tv, now in Bloomberg and Fox Business. Thanks, Brian. Thanks to everyone.